Generative AI detox - how did it go?
Summary
I undertook a short, generative AI detox. It helped me reinforce that while LLMs aren’t indispensable, I don’t mind using them for boring work. I am, however, going to militantly avoid generative AI in activities that I consider to be my superpowers.
A couple of weeks back, I started a week-long generative AI detox. I survived the week; can you believe it? I’ve also had a week to reflect on the detox and life before and after that week. So, here’s a short report on how it went.
What was good
Cutting off generative AI was excellent in so many ways. First, it reinforced for me that if LLMs disappear tomorrow or become unaffordable, I can do without them. Neither my job nor my life would be significantly worse without these tools.
Second, it helped me appreciate the less seductive AI that’s been around for donkeys – text to speech and speech to text, object detection, denoising and demosaicing, etc – I don’t want these tools to go away.
Third, I enjoyed using simpler tools. For example, I stopped using Google Search and the atrocious AI overviews and chose Kagi instead. Kagi is an incredible search platform. The results are better than Google's, and I get to fine-tune them to my preferences and filters. No sneaky algorithm second-guessing what I want, no one trying to sell me stuff. None of that nonsense.
Fourth, and most importantly, I made it a point not to consume any AI news, not even Ed Zitron and Cal Newport, whom I adore. AI reporting is stressful to keep up with. There’s the AI booster newsfeed, which, as Newport characterises, is full of vibe reporting, digital ick and faux astonishment. But even hearing from sceptics can be tiring. Not because they’re wrong, but because they have no option but to react to the boosters constantly. Shutting off this newsfeed did wonders for my mental health during that detox week.
What was hard
I won’t deny that I am getting used to having generative AI tools do some of the boring work I hate to do by hand. For example, I had to create an affidavit for my mom using a printed template. I could have handed that print to an LLM, but I used OCR instead. It was fiddly, but it wasn’t the end of the world.
I’ve often used YouTube as a source for podcasts, and when I’m using generative AI, I get a textual summary of the episode to know what it's about, before I give it a whole-hearted listen. I couldn’t use this workflow, and I missed it.
I often use LLMs to generate images with a short shelf life, and it was annoying to do these by hand, but I didn’t mind the fine-grained control I got back in the process. There was a time I went to Gemini just out of habit, and I even generated an image, but I ended up not using output.
The hardest part was avoiding generative AI for work. I had to build an agent on Glean, as part of my work, so I couldn’t help but break out of detox. Another time, a colleague needed help with Rovo, and I couldn’t just sit back and ignore them, so I broke out of detox again. I could count these instances on one hand, but it’s worth noting how much LLMs have penetrated the workplace.
What I’m taking away
In the week after the detox, I’ve resumed using LLMs and here’s my takeaway. I like LLMs for doing tasks that feel like the knowledge work equivalents of root canals. Transcribing and summarising meeting notes, generating a weekly report, or even copyediting my own writing. I’m quite happy to delegate them to word-guessing machines as long as they’re affordable and accessible. Since most LLM-powered tools employ looping these days, they’re in the ballpark of the right answer. So, I don’t mind spinning the probabilistic roulette for the grunt work we must often do.
That said, there are skills that I consider my superpowers. I enjoy practising them, and since I’ve been at them for many years now, I’m already better at them than most people. Since LLMs only extrapolate patterns from existing content, they’re prone to the garbage-in, garbage-out phenomenon. I see no point in lowering my skills to the common denominator outputs of generative AI. I’ll keep my edge, thank you very much!
As a technologist, especially one who works in a tech services firm, I need to be conversant with generative AI until the bubble bursts (if it ever will). I can’t renounce generative AI yet, but if anything, the detox has helped me reinforce my belief that we must balance our wide-eyed wonder for technology with narrow-eyed scepticism.